SpeechSense: An Interactive Sensor Platform for Speech Therapy

这一阶段将开发SpeechSense™：一种用于运动语言治疗的交互式传感器平台 在美国，超过1000万人的语音交流受到障碍。对这些人的关怀是 主要是在临床上使用知觉量表-这存在评分者之间的低可靠性-或者 用于量化语音声学测量的复杂设备--这很容易受到对话的影响 噪音，因此仍然局限于受控的脚本朗诵。因此，量化措施 在日常生活的自然对话互动中评估言语障碍对于言语是不可用的- 语言病理学家(SLP)，防止他们获得关于存在、严重程度、 和功能障碍的影响，并限制治疗成果从临床带入日常生活 生活。为了满足这一需求，我们的人体测量技术专家团队正在与领先的电机公司合作 波士顿大学的语音研究人员和SLP开发一种新型的混合式声学-加速度计传感器 与软件配对，用于自动降噪和导出发声和发音测量 评估自然的对话语言。声音信号可以提供健壮的语音清晰度测量 但很难在环境噪音或其他扬声器的声音中隔离发声措施；而加速计 当获得语音的声学测量时，录音对这种噪声更加健壮，但仍然不知道 讲话的发音语境。因此，结合这两种传感器模式提供了独特的机会 在自然的对话互动中获得发声和发音的测量。我们的第一阶段计划将定制 设计微控制器、软件和固件，将声学麦克风和加速度计集成到 单一的、颈部佩戴的传感器，将用于从运动能力低下的患者那里获取语音数据语料库 帕金森氏病(PD)引起的构音障碍在有或没有各种来源的对话活动中 噪音。利用这些数据，我们将开发一系列的数据融合、模式识别和信号处理 用于自动区分和减少感兴趣的噪声源以获得临床测量的算法 语音功能(如基频、发音元音间距、语速、声门下压力、 和其他)，根据黄金标准的临床程序验证它们，并在 不同的噪音条件。传感器原型和测量软件将由我们的SLP团队进行测试 对伴有运动性构音障碍的帕金森病患者进行测试，以证明言语感觉™为 基于积极的SLP和患者自我报告的可用性脚本和对话评估， 可接受性和感知价值。这一概念验证将为开发第二阶段预 带有实时算法和移动软件的商业原型，将为SLP提供新的工具 加强语音治疗，改进临床评估，并在自然对话期间监测治疗情况 在个人最需要提高生活质量的情况下进行互动。

This Phase I SBIR will develop SpeechSense™: An Interactive Sensor Platform for Speech Therapy of motor speech disorders impeding vocal communication for over 10M individuals in the US. Care for these individuals is primarily done in the clinic using either perceptual scales–which suffer from low inter-rater reliability–or sophisticated equipment for quantifying acoustic measures of speech—which is susceptible to conversational noise and therefore remains limited to controlled scripted recitations. As a result, quantitative measures for evaluating speech impairments during natural conversational interactions of daily life are unavailable to speech- language pathologists (SLPs), preventing them from obtaining a complete description of the presence, severity, and functional impact of a disorder, and limiting the carryover of therapeutic gains from the clinic into daily life. To meet this need, our team of experts in human measurement technology is partnering with leading motor speech researchers and SLPs at Boston University to develop a novel hybrid acoustic-accelerometer sensor paired to software for automated noise mitigation and derivation of vocal and articulatory measures for assessing natural conversational speech. Acoustic signals can provide robust articulatory measures of speech but struggle to isolate vocal measures amid ambient noise or the sound of other speakers; while accelerometer recordings are more robust to such noise when obtaining vocal measures of speech, but remain agnostic to the articulatory context of speech. Combining both sensor modalities therefore offers the unique opportunity to obtain vocal and articulatory measures during natural conversational interactions. Our Phase I plan will custom design a microcontroller, software, and firmware to integrate an acoustic microphone and accelerometer into a single, neck-worn sensor, which will be used to acquire a corpus of speech data from patients with hypokinetic dysarthria from Parkinson’s disease (PD) during conversational activities with and without various sources of noise. Using these data, we will develop a series of data fusion, pattern recognition, and signal processing algorithms to autonomously discriminate and mitigate noise sources of interest for deriving clinical measures of speech function (such as fundamental frequency, articulatory vowel space, speech rate, subglottal pressure, and others), validate them with respect to gold-standard clinical procedures, and test their reliability under different noise conditions. The sensor prototype and measurement software will be tested by our team of SLPs on PD patients with hypokinetic dysarthria to demonstrate that SpeechSense™ provides a feasible modality for both scripted and conversational assessment based on positive SLP and patient self-reports for usability, acceptability, and perceived value. This proof of concept will lay the foundation for developing a Phase II pre- commercial prototype with real-time algorithms and mobile software that will provide a new tool for SLPs to augment voice therapy, improve clinical assessment, and monitor treatment during natural conversational interactions where individuals experience the greatest need to improve quality of life.